Re: RMAN or Hot Backup

From: joel garry <>
Date: Mon, 13 Apr 2009 12:25:36 -0700 (PDT)
Message-ID: <>

On Apr 10, 7:45 pm, "Bob Jones" <em..._at_me.not> wrote:
> "Michael Austin" <> wrote in message
> news:fpSDl.26820$
> > Bob Jones wrote:
> >> "joel garry" <> wrote in message
> >>
> >> On Apr 9, 5:07 pm, "Bob Jones" <em..._at_me.not> wrote:
> >>> "jgar the jorrible" <> wrote in
> >>>
> >>> On Mar 27, 5:15 pm, "Bob Jones" <em..._at_me.not> wrote:
> >>>>> So yes, it is DBA basics. I really have not done too many Hot
> >>>>> Backups, which is why I was asking what others opinions and experience
> >>>>> with both are.
> >>>> I have no idea why you guys keep referring to non-RMAN backup as "hot
> >>>> backup". That is just wrong terminology.
> >>> - We are referring to the copying of Oracle files while the db is
> >>> - running. This results in an inconsistent set of data files, which can
> >>> - be made consistent on recovery by the application of redo. However,
> >>> - if the data files are not placed into backup mode before copying,
> >>> - there may not be enough information in redo to make them consistent.
> >>> - RMAN inconsistent backups are also hot backups, but RMAN is smarter
> >>> - than operating system utilities and can copy the blocks in the data
> >>> - files in a way that avoids the problems of redo, as well as avoiding
> >>> - suspending the system.
> >>> I am not sure what you meant by avoiding redo problems and suspending
> >>> the
> >>> system. I see neither being a problem with non-RMAN backups.
> >> - 1.  More redo is generated.  This can be an issue on a system that
> >> - generates lots of redo to begin with.
> >> Just make sure I hear this correctly. Non-RMAN backups cause more redo to
> >> be generated?
> >> - 2.  It becomes common for tablespaces to be left in hot backup mode,
> >> - then people ask things like "why am I generating so much redo?"
> >> Still not understanding. Why would it generate more redo logs?
> >> - 3.  As Michael pointed out, suspend can be a problem on an unquiet
> >> - system.
> >> User-managed backups absolutely do not require system to be suspended. We
> >> were doing backups long before RMAN even existed. Never once we had to
> >> suspend the system.
> > Bob, you miss understood this...  This is for a particular Storage Array
> > Vendor "copy" solution called a BCV (Business Continuity Volume) copy
> > where the database is put into hot-backup mode, alter system suspend,
> > split the "mirror", alter system resume, undo hot-backup mode.
> > Having done these splits without the suspend mode you risk ORA-600s with
> > corrupted undo segments. Been there, done that.
> Who said this is for your kind of backup in particular? We are talking about
> RMAN vs non-RMAN backup. Not using RMAN does not mean you have to suspend
> the system to do a backup.

This is simply highlighting what kinds of variants there are of backups. Michael's viewpoint is obviously larger databases, which often have special requirements. My viewpoint tends towards midsized databases, which have more general backup requirements, but can suffer from both being treated as non-oracle databases, as well as some of these bizarro solutions that require suspend. I may happen to be stuck with one of these bizarro solutions if I don't look at these things, hence the thread-drift and my interest in the details of what people have seen.

And given the broad thread to start, it may not even be a drift.

Yes, you may use cold backups and not have to worry about backup mode and suspend. However, you might consider how 10g and 11g are delivered and configured these days, before you assume nothing is happening out of "normal" business hours. In particular, look at the maintenance windows, and think about why there might be maintenance windows. IMNSHO, we are already at the point where cold backups are considered outside normal practice and need special justification. "We've always done that" is not a good reason after upgrading.

It was not so long ago I was saying intermittent cold backups were a good idea, I no longer think that, and probably should have not thought that from long ago. 10+ years ago I worked on a solution where the volume manager went bongos (solaris/veritas), and got a bit paranoid.


-- is bogus.
Received on Mon Apr 13 2009 - 14:25:36 CDT

Original text of this message